[RLlib] Add Segmentation Buffer for DT #27829
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
charlesjsun
requested review from
avnishn,
ArturNiederfahrenhorst,
smorad,
maxpumperla,
kouroshHakha and
krfricke
as code owners
charlesjsun
changed the title
charlesjsun
commented
Aug 12, 2022
| # decision transformer | ||
| RETURNS_TO_GO = "returns_to_go" | ||
| ATTENTION_MASKS = "attention_masks" | ||
|
|
There was a problem hiding this comment.
Arguments could be made for putting it here, or putting it elsewhere like in dt.py which doesn't exist yet.
Signed-off-by: Charles Sun <[email protected]>
charlesjsun
force-pushed
the
dt_segmentation_buffer
branch
from
9f8ce0b
to
d6d446c
Compare
charlesjsun
changed the title
avnishn
reviewed
Aug 15, 2022
| # TODO: sample proportional to episode length | ||
| # Sample a random episode from the buffer and then sample a random | ||
| # segment from that episode. | ||
| buffer_ind = np.random.randint(0, len(self._buffer)) |
There was a problem hiding this comment.
this is not a seedable call. Can you instantiate your class with a numpy rng and set the seed of the rng using the seed in the global config?
|
|
||
| obs = episode[SampleBatch.OBS][si:ei] | ||
| actions = episode[SampleBatch.ACTIONS][si:ei] | ||
| # Note that returns-to-go needs an extra elem as the target for the last action. |
| [returns_to_go, np.zeros((1, 1), dtype=returns_to_go.dtype)], axis=0 | ||
| ) | ||
|
|
||
| # Front-pad if at beginning of rollout. |
There was a problem hiding this comment.
add note that this is about inference :)
| offset = min(self.max_seq_len, ep_len) | ||
| # We allow si to be negative (for now) because we want segments that only | ||
| # contains the first few transitions (and padd the rest), | ||
| # for example [0, 0, 0, 0, 0, 0, R0, s0, a0]. |
| # add to buffer and check that only last one is kept (due to replacement) | ||
| buffer.add(batch) | ||
|
|
||
| assert len(_get_internal_buffer(buffer)) == 1 |
| # decision transformer | ||
| RETURNS_TO_GO = "returns_to_go" | ||
| ATTENTION_MASKS = "attention_masks" | ||
|
|
Signed-off-by: Charles Sun <[email protected]>
kouroshHakha
requested changes
Aug 16, 2022
| # TODO: sample proportional to episode length | ||
| # Sample a random episode from the buffer and then sample a random | ||
| # segment from that episode. | ||
| buffer_ind = np.random.randint(0, len(self._buffer)) |
Signed-off-by: Charles Sun <[email protected]>
Signed-off-by: Charles Sun <[email protected]>
Signed-off-by: Charles Sun <[email protected]>
Signed-off-by: Charles Sun <[email protected]>
kouroshHakha
approved these changes
Aug 16, 2022
|
@richardliaw This can be merged. It passes all the tests. |
Stefan-1313
pushed a commit
to Stefan-1313/ray_mod
that referenced
this pull request
Aug 18, 2022
Signed-off-by: Stefan van der Kleij <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Signed-off-by: Charles Sun [email protected]
Why are these changes needed?
Added the SegmentationBuffer that DT (Decision Transformer) needs.
Related issue number
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.